Node.js Streams: tout ce que vous devez savoir

Mise à jour: Cet article fait maintenant partie de mon livre «Node.js Beyond The Basics».

Lisez la version mise à jour de ce contenu et plus d'informations sur Node sur jscomplete.com/node-beyond-basics .

Les flux Node.js ont la réputation d'être difficiles à utiliser et encore plus difficiles à comprendre. Eh bien, j'ai de bonnes nouvelles pour vous - ce n'est plus le cas.

Au fil des ans, les développeurs ont créé de nombreux packages dans le seul but de faciliter le travail avec les flux. Mais dans cet article, je vais me concentrer sur l'API de flux native Node.js.

"Les flux sont la meilleure idée et la plus incomprise de Node."

- Dominic Tarr

Que sont exactement les flux?

Les flux sont des collections de données - tout comme des tableaux ou des chaînes. La différence est que les flux peuvent ne pas être disponibles en même temps et qu'ils n'ont pas besoin de tenir en mémoire. Cela rend les flux vraiment puissants lorsque vous travaillez avec de grandes quantités de données ou des données provenant d'une source externe, un morceau à la fois.

Cependant, les flux ne consistent pas uniquement à travailler avec des données massives. Ils nous donnent également le pouvoir de la composabilité dans notre code. Tout comme nous pouvons composer de puissantes commandes Linux en envoyant d'autres commandes Linux plus petites, nous pouvons faire exactement la même chose dans Node avec des flux.

const grep = ... // A stream for the grep output const wc = ... // A stream for the wc input grep.pipe(wc)

De nombreux modules intégrés dans Node implémentent l'interface de streaming:

La liste ci-dessus contient quelques exemples d'objets Node.js natifs qui sont également des flux lisibles et inscriptibles. Certains de ces objets sont à la fois des flux lisibles et inscriptibles, comme les sockets TCP, zlib et les flux cryptographiques.

Notez que les objets sont également étroitement liés. Alors qu'une réponse HTTP est un flux lisible sur le client, c'est un flux accessible en écriture sur le serveur. C'est parce que dans le cas HTTP, nous lisons essentiellement à partir d'un objet ( http.IncomingMessage) et écrivons dans l'autre ( http.ServerResponse).

A noter également la façon dont les stdioflux ( stdin, stdout, stderr) ont les types de flux inverse en ce qui concerne les processus enfants. Cela permet un moyen très simple de diriger vers et depuis ces flux à partir des flux de processus principaux stdio.

Un exemple pratique de streams

La théorie est excellente, mais souvent pas convaincante à 100%. Voyons un exemple illustrant la différence que les flux peuvent faire dans le code en ce qui concerne la consommation de mémoire.

Créons d'abord un gros fichier:

const fs = require('fs'); const file = fs.createWriteStream('./big.file'); for(let i=0; i<= 1e6; i++) { file.write('Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.\n'); } file.end();

Regardez ce que j'ai utilisé pour créer ce gros fichier. Un flux inscriptible!

Le fsmodule peut être utilisé pour lire et écrire dans des fichiers à l'aide d'une interface de flux. Dans l'exemple ci-dessus, nous y écrivons big.filevia un flux inscriptible de 1 million de lignes avec une boucle.

L'exécution du script ci-dessus génère un fichier d'environ 400 Mo.

Voici un serveur Web Node simple conçu pour servir exclusivement big.file:

const fs = require('fs'); const server = require('http').createServer(); server.on('request', (req, res) => { fs.readFile('./big.file', (err, data) => { if (err) throw err; res.end(data); }); }); server.listen(8000);

Lorsque le serveur reçoit une requête, il va servir le grand fichier en utilisant la méthode asynchrone, fs.readFile. Mais bon, ce n'est pas comme si nous bloquions la boucle d'événements ou quoi que ce soit. Tout est génial, non? Droite?

Eh bien, voyons ce qui se passe lorsque nous exécutons le serveur, nous y connectons et surveillons la mémoire tout en le faisant.

Lorsque j'ai exécuté le serveur, il a commencé avec une quantité normale de mémoire, 8,7 Mo:

Puis je me suis connecté au serveur. Notez ce qui est arrivé à la mémoire consommée:

Wow - la consommation de mémoire est passée à 434,8 Mo.

Nous mettons essentiellement tout le big.filecontenu en mémoire avant de l'écrire dans l'objet de réponse. C'est très inefficace.

L'objet de réponse HTTP ( resdans le code ci-dessus) est également un flux accessible en écriture. Cela signifie que si nous avons un flux lisible qui représente le contenu de big.file, nous pouvons simplement les diriger l'un sur l'autre et obtenir presque le même résultat sans consommer ~ 400 Mo de mémoire.

Le fsmodule de Node peut nous donner un flux lisible pour n'importe quel fichier utilisant la createReadStreamméthode. Nous pouvons diriger cela vers l'objet de réponse:

const fs = require('fs'); const server = require('http').createServer(); server.on('request', (req, res) => { const src = fs.createReadStream('./big.file'); src.pipe(res); }); server.listen(8000);

Maintenant, lorsque vous vous connectez à ce serveur, une chose magique se produit (regardez la consommation de mémoire):

Que ce passe-t-il?

Lorsqu'un client demande ce gros fichier, nous le diffusons un morceau à la fois, ce qui signifie que nous ne le mettons pas du tout en mémoire tampon. L'utilisation de la mémoire a augmenté d'environ 25 Mo et c'est tout.

Vous pouvez pousser cet exemple à ses limites. Régénérez le big.fileavec cinq millions de lignes au lieu d'un million, ce qui porterait le fichier à bien plus de 2 Go, et c'est en fait plus grand que la limite de tampon par défaut dans Node.

Si vous essayez de servir ce fichier en utilisant fs.readFile, vous ne pouvez tout simplement pas, par défaut (vous pouvez modifier les limites). Mais avec fs.createReadStream, il n'y a aucun problème à diffuser 2 Go de données vers le demandeur et, mieux encore, l'utilisation de la mémoire du processus sera à peu près la même.

Prêt à apprendre les flux maintenant?

Cet article est une rédaction d'une partie de mon cours Pluralsight sur Node.js. J'y couvre un contenu similaire au format vidéo.

Flux 101

Il existe quatre types de flux fondamentaux dans Node.js: les flux lisibles, inscriptibles, duplex et de transformation.

  • Un flux lisible est une abstraction pour une source à partir de laquelle des données peuvent être consommées. Un exemple de cela est la fs.createReadStreamméthode.
  • Un flux inscriptible est une abstraction pour une destination sur laquelle les données peuvent être écrites. Un exemple de cela est la fs.createWriteStreamméthode.
  • Un flux duplex est à la fois lisible et inscriptible. Un exemple de cela est une socket TCP.
  • Un flux de transformation est essentiellement un flux duplex qui peut être utilisé pour modifier ou transformer les données lors de leur écriture et de leur lecture. Un exemple de cela est le zlib.createGzipflux pour compresser les données à l'aide de gzip. Vous pouvez considérer un flux de transformation comme une fonction où l'entrée est la partie de flux accessible en écriture et la sortie est la partie de flux lisible. Vous pouvez également entendre les flux de transformation appelés " flux via les flux ".

Tous les flux sont des instances de EventEmitter. Ils émettent des événements qui peuvent être utilisés pour lire et écrire des données. Cependant, nous pouvons consommer des données de flux de manière plus simple en utilisant la pipeméthode.

La méthode du tuyau

Voici la ligne magique dont vous devez vous souvenir:

readableSrc.pipe(writableDest)

In this simple line, we’re piping the output of a readable stream — the source of data, as the input of a writable stream — the destination. The source has to be a readable stream and the destination has to be a writable one. Of course, they can both be duplex/transform streams as well. In fact, if we’re piping into a duplex stream, we can chain pipe calls just like we do in Linux:

readableSrc .pipe(transformStream1) .pipe(transformStream2) .pipe(finalWrtitableDest)

The pipe method returns the destination stream, which enabled us to do the chaining above. For streams a (readable), b and c (duplex), and d (writable), we can:

a.pipe(b).pipe(c).pipe(d) # Which is equivalent to: a.pipe(b) b.pipe(c) c.pipe(d) # Which, in Linux, is equivalent to: $ a | b | c | d

The pipe method is the easiest way to consume streams. It’s generally recommended to either use the pipe method or consume streams with events, but avoid mixing these two. Usually when you’re using the pipe method you don’t need to use events, but if you need to consume the streams in more custom ways, events would be the way to go.

Stream events

Beside reading from a readable stream source and writing to a writable destination, the pipe method automatically manages a few things along the way. For example, it handles errors, end-of-files, and the cases when one stream is slower or faster than the other.

However, streams can also be consumed with events directly. Here’s the simplified event-equivalent code of what the pipe method mainly does to read and write data:

# readable.pipe(writable) readable.on('data', (chunk) => { writable.write(chunk); }); readable.on('end', () => { writable.end(); });

Here’s a list of the important events and functions that can be used with readable and writable streams:

The events and functions are somehow related because they are usually used together.

The most important events on a readable stream are:

  • The data event, which is emitted whenever the stream passes a chunk of data to the consumer
  • The end event, which is emitted when there is no more data to be consumed from the stream.

The most important events on a writable stream are:

  • The drain event, which is a signal that the writable stream can receive more data.
  • The finish event, which is emitted when all data has been flushed to the underlying system.

Events and functions can be combined to make for a custom and optimized use of streams. To consume a readable stream, we can use the pipe/unpipe methods, or the read/unshift/resume methods. To consume a writable stream, we can make it the destination of pipe/unpipe, or just write to it with the write method and call the end method when we’re done.

Paused and Flowing Modes of Readable Streams

Readable streams have two main modes that affect the way we can consume them:

  • They can be either in the paused mode
  • Or in the flowing mode

Those modes are sometimes referred to as pull and push modes.

All readable streams start in the paused mode by default but they can be easily switched to flowing and back to paused when needed. Sometimes, the switching happens automatically.

When a readable stream is in the paused mode, we can use the read() method to read from the stream on demand, however, for a readable stream in the flowing mode, the data is continuously flowing and we have to listen to events to consume it.

In the flowing mode, data can actually be lost if no consumers are available to handle it. This is why, when we have a readable stream in flowing mode, we need a data event handler. In fact, just adding a data event handler switches a paused stream into flowing mode and removing the data event handler switches the stream back to paused mode. Some of this is done for backward compatibility with the older Node streams interface.

To manually switch between these two stream modes, you can use the resume() and pause() methods.

When consuming readable streams using the pipe method, we don’t have to worry about these modes as pipe manages them automatically.

Implementing Streams

When we talk about streams in Node.js, there are two main different tasks:

  • The task of implementing the streams.
  • The task of consuming them.

So far we’ve been talking about only consuming streams. Let’s implement some!

Stream implementers are usually the ones who require the stream module.

Implementing a Writable Stream

To implement a writable stream, we need to to use the Writable constructor from the stream module.

const { Writable } = require('stream');

We can implement a writable stream in many ways. We can, for example, extend the Writable constructor if we want

class myWritableStream extends Writable { }

However, I prefer the simpler constructor approach. We just create an object from the Writable constructor and pass it a number of options. The only required option is a write function which exposes the chunk of data to be written.

const { Writable } = require('stream'); const outStream = new Writable({ write(chunk, encoding, callback) { console.log(chunk.toString()); callback(); } }); process.stdin.pipe(outStream);

This write method takes three arguments.

  • The chunk is usually a buffer unless we configure the stream differently.
  • The encoding argument is needed in that case, but usually we can ignore it.
  • The callback is a function that we need to call after we’re done processing the data chunk. It’s what signals whether the write was successful or not. To signal a failure, call the callback with an error object.

In outStream, we simply console.log the chunk as a string and call the callback after that without an error to indicate success. This is a very simple and probably not so useful echo stream. It will echo back anything it receives.

To consume this stream, we can simply use it with process.stdin, which is a readable stream, so we can just pipe process.stdin into our outStream.

When we run the code above, anything we type into process.stdin will be echoed back using the outStreamconsole.log line.

This is not a very useful stream to implement because it’s actually already implemented and built-in. This is very much equivalent to process.stdout. We can just pipe stdin into stdout and we’ll get the exact same echo feature with this single line:

process.stdin.pipe(process.stdout);

Implement a Readable Stream

To implement a readable stream, we require the Readable interface, and construct an object from it, and implement a read() method in the stream’s configuration parameter:

const { Readable } = require('stream'); const inStream = new Readable({ read() {} });

There is a simple way to implement readable streams. We can just directly push the data that we want the consumers to consume.

const { Readable } = require('stream'); const inStream = new Readable({ read() {} }); inStream.push('ABCDEFGHIJKLM'); inStream.push('NOPQRSTUVWXYZ'); inStream.push(null); // No more data inStream.pipe(process.stdout);

When we push a null object, that means we want to signal that the stream does not have any more data.

To consume this simple readable stream, we can simply pipe it into the writable stream process.stdout.

When we run the code above, we’ll be reading all the data from inStream and echoing it to the standard out. Very simple, but also not very efficient.

We’re basically pushing all the data in the stream before piping it to process.stdout. The much better way is to push data on demand, when a consumer asks for it. We can do that by implementing the read() method in the configuration object:

const inStream = new Readable({ read(size) { // there is a demand on the data... Someone wants to read it. } });

When the read method is called on a readable stream, the implementation can push partial data to the queue. For example, we can push one letter at a time, starting with character code 65 (which represents A), and incrementing that on every push:

const inStream = new Readable({ read(size) { this.push(String.fromCharCode(this.currentCharCode++)); if (this.currentCharCode > 90) { this.push(null); } } }); inStream.currentCharCode = 65; inStream.pipe(process.stdout);

While the consumer is reading a readable stream, the read method will continue to fire, and we’ll push more letters. We need to stop this cycle somewhere, and that’s why an if statement to push null when the currentCharCode is greater than 90 (which represents Z).

This code is equivalent to the simpler one we started with but now we’re pushing data on demand when the consumer asks for it. You should always do that.

Implementing Duplex/Transform Streams

With Duplex streams, we can implement both readable and writable streams with the same object. It’s as if we inherit from both interfaces.

Here’s an example duplex stream that combines the two writable and readable examples implemented above:

const { Duplex } = require('stream'); const inoutStream = new Duplex({ write(chunk, encoding, callback) { console.log(chunk.toString()); callback(); }, read(size) { this.push(String.fromCharCode(this.currentCharCode++)); if (this.currentCharCode > 90) { this.push(null); } } }); inoutStream.currentCharCode = 65; process.stdin.pipe(inoutStream).pipe(process.stdout);

By combining the methods, we can use this duplex stream to read the letters from A to Z and we can also use it for its echo feature. We pipe the readable stdin stream into this duplex stream to use the echo feature and we pipe the duplex stream itself into the writable stdout stream to see the letters A through Z.

It’s important to understand that the readable and writable sides of a duplex stream operate completely independently from one another. This is merely a grouping of two features into an object.

A transform stream is the more interesting duplex stream because its output is computed from its input.

For a transform stream, we don’t have to implement the read or write methods, we only need to implement a transform method, which combines both of them. It has the signature of the write method and we can use it to push data as well.

Here’s a simple transform stream which echoes back anything you type into it after transforming it to upper case format:

const { Transform } = require('stream'); const upperCaseTr = new Transform({ transform(chunk, encoding, callback) { this.push(chunk.toString().toUpperCase()); callback(); } }); process.stdin.pipe(upperCaseTr).pipe(process.stdout);

In this transform stream, which we’re consuming exactly like the previous duplex stream example, we only implemented a transform() method. In that method, we convert the chunk into its upper case version and then push that version as the readable part.

Streams Object Mode

By default, streams expect Buffer/String values. There is an objectMode flag that we can set to have the stream accept any JavaScript object.

Here’s a simple example to demonstrate that. The following combination of transform streams makes for a feature to map a string of comma-separated values into a JavaScript object. So “a,b,c,d” becomes {a: b, c: d}.

const { Transform } = require('stream'); const commaSplitter = new Transform({ readableObjectMode: true, transform(chunk, encoding, callback) { this.push(chunk.toString().trim().split(',')); callback(); } }); const arrayToObject = new Transform({ readableObjectMode: true, writableObjectMode: true, transform(chunk, encoding, callback) { const obj = {}; for(let i=0; i < chunk.length; i+=2) { obj[chunk[i]] = chunk[i+1]; } this.push(obj); callback(); } }); const objectToString = new Transform({ writableObjectMode: true, transform(chunk, encoding, callback) { this.push(JSON.stringify(chunk) + '\n'); callback(); } }); process.stdin .pipe(commaSplitter) .pipe(arrayToObject) .pipe(objectToString) .pipe(process.stdout)

We pass the input string (for example, “a,b,c,d”) through commaSplitter which pushes an array as its readable data ([“a”, “b”, “c”, “d”]). Adding the readableObjectMode flag on that stream is necessary because we’re pushing an object there, not a string.

We then take the array and pipe it into the arrayToObject stream. We need a writableObjectMode flag to make that stream accept an object. It’ll also push an object (the input array mapped into an object) and that’s why we also needed the readableObjectMode flag there as well. The last objectToString stream accepts an object but pushes out a string, and that’s why we only needed a writableObjectMode flag there. The readable part is a normal string (the stringified object).

Node’s built-in transform streams

Node has a few very useful built-in transform streams. Namely, the zlib and crypto streams.

Here’s an example that uses the zlib.createGzip() stream combined with the fs readable/writable streams to create a file-compression script:

const fs = require('fs'); const zlib = require('zlib'); const file = process.argv[2]; fs.createReadStream(file) .pipe(zlib.createGzip()) .pipe(fs.createWriteStream(file + '.gz'));

You can use this script to gzip any file you pass as the argument. We’re piping a readable stream for that file into the zlib built-in transform stream and then into a writable stream for the new gzipped file. Simple.

The cool thing about using pipes is that we can actually combine them with events if we need to. Say, for example, I want the user to see a progress indicator while the script is working and a “Done” message when the script is done. Since the pipe method returns the destination stream, we can chain the registration of events handlers as well:

const fs = require('fs'); const zlib = require('zlib'); const file = process.argv[2]; fs.createReadStream(file) .pipe(zlib.createGzip()) .on('data', () => process.stdout.write('.')) .pipe(fs.createWriteStream(file + '.zz')) .on('finish', () => console.log('Done'));

So with the pipe method, we get to easily consume streams, but we can still further customize our interaction with those streams using events where needed.

What’s great about the pipe method though is that we can use it to compose our program piece by piece, in a much readable way. For example, instead of listening to the data event above, we can simply create a transform stream to report progress, and replace the .on() call with another .pipe() call:

const fs = require('fs'); const zlib = require('zlib'); const file = process.argv[2]; const { Transform } = require('stream'); const reportProgress = new Transform({ transform(chunk, encoding, callback) { process.stdout.write('.'); callback(null, chunk); } }); fs.createReadStream(file) .pipe(zlib.createGzip()) .pipe(reportProgress) .pipe(fs.createWriteStream(file + '.zz')) .on('finish', () => console.log('Done'));

This reportProgress stream is a simple pass-through stream, but it reports the progress to standard out as well. Note how I used the second argument in the callback() function to push the data inside the transform() method. This is equivalent to pushing the data first.

The applications of combining streams are endless. For example, if we need to encrypt the file before or after we gzip it, all we need to do is pipe another transform stream in that exact order that we needed. We can use Node’s crypto module for that:

const crypto = require('crypto'); // ... fs.createReadStream(file) .pipe(zlib.createGzip()) .pipe(crypto.createCipher('aes192', 'a_secret')) .pipe(reportProgress) .pipe(fs.createWriteStream(file + '.zz')) .on('finish', () => console.log('Done'));

The script above compresses and then encrypts the passed file and only those who have the secret can use the outputted file. We can’t unzip this file with the normal unzip utilities because it’s encrypted.

To actually be able to unzip anything zipped with the script above, we need to use the opposite streams for crypto and zlib in a reverse order, which is simple:

fs.createReadStream(file) .pipe(crypto.createDecipher('aes192', 'a_secret')) .pipe(zlib.createGunzip()) .pipe(reportProgress) .pipe(fs.createWriteStream(file.slice(0, -3))) .on('finish', () => console.log('Done'));

Assuming the passed file is the compressed version, the code above will create a read stream from that, pipe it into the crypto createDecipher() stream (using the same secret), pipe the output of that into the zlib createGunzip() stream, and then write things out back to a file without the extension part.

That’s all I have for this topic. Thanks for reading! Until next time!

Learning React or Node? Checkout my books:

  • Learn React.js by Building Games
  • Node.js Beyond the Basics